0
srRUGTURA1
MECHANICS0
APPLIED MAThEMA1ICS I $ILL Sf1
Tøchnisce
THE IMPULSE RESPONSE FUNCTION AND SHIP MOTIONS
W. L Cummins
AERODYNAMICS
For presentation at the
SYMPOSIUM ON SUP TREQRY
Institut fUr Schiffbau der UniverSitt Hamburg Hamburg, Germany
25 - 27 January 1962
THE IMPULSE RESPONSE PUNCION AND SHIP MOTIONS
by
W. E. Cunnnins
For presentation at the
SYMPOSIUM ON SNIP THEORY
Institut für Schiffbau der Universität Hamburg Hamburg., Germany
Just over a decade ago, Weinbium and St. Denis1 presented a compre-hensive review of the state of knowledge at the end of what we may call
the "classicalt' period in research on seakeeping. Soon after, St. Denis
and Pierson3 opened the "modern" period (some would prefer to call it
the "statistical" period) The studies of the former period were
pri-marily concerned with sinusoidal responses to sinusoidal waves, but the introduction of spectral techniques opened the door for the discusion of responses to random waves, both. long and short crested. The construction of the spectral theory on regular wave theory as a foundation deligbted
us all,, as it presented an apparent justification for the admittedly
artificial studies of the "classical"pperiod.
The activity during this last decade has been spectacular, with fiv-e major and many minor facilities for seakeeping research being opened.
Hundreds of models have been tested, many full scale trials have been run, and there has even been some real growth in our knowledge of the subject.
In particular, the spectral tool has been sharpened and tempered by the
mpricistS, and the analysts have made important advances with the rather
frightful boundary value problem. In fact, we have all been forging ahead so rapidly, that we appear to have forgotten that we are wearing a shoe which doesn't quite fit. The occasional pain from a misplaced toe is
ignored in our general enthusiasm for progress.
The "shoe" to which I refer is our inathei4tical model, the.forced
representation of the ship response by a system of secd order differential
equations. The shoe is.squeezed on, with no regard for the shape of the foot. The inadequacy of the shoe is evident in the distortions it must take if it is to be worn at all. I am referring, of course, to the fre-quency dependent coefficients which permit the mathematical model to fit the physical model, (if the excitation is purely sinusoidal, that is).
But what happens when we dont have a well defined frequency? The
mathematical model becomes almost meaningless. True, a Fourier analysis
of -the exciting force (or encountered wave) permits the modeltô be
retained, but physical realit is almost lost-in the infinit' of equations
req.iired to represent the motion. The all important intuition of the engineer Ic heavily encumbered with this grotesque baggage.
Let us consider this mathematical model briefly, and restrict ourselves
to a single degree of freedom. To becompletely fair, iet.us consider a
pure, sinusoidal oscilation. The forcing function (if the sytem is linear) will be sinusoidal, and can be b4ken into two components, one in phase with
the displacement and one 900 out of phase. We further divide the in-phase component into a. restoring force, proportional to the displacement, and a remainder. The latter, we call the inertial force,
and treat it-as if it
were proportional to the instantaneous acceleration. The out of phase
component, which provides all the damping, we treat as if it were proportional
to the instantaneous velocity.
We can now write an equation, which has the appearance of a differential
equation, relating these various quantities:
+ b(w)cc + c.(w)x- F sin w(t-i-)
But a differential equation is supposed to relate the instantaneous values of the function-iàvoigêd. If the periodic motl.on continues, this condition
ssatisfied.Of, course, it could just as well be satisfied by the-equation
bk + (c-auP)x = f(t)
or, more generally
(a+d) + bx +
(cZIuP)x
= f(t)where d is arbitrary. These are all.equally valid models.
One of them sto be preferred only if it truly relates the displacement and its first
and second derivatives to the excitation. But suppose f(t) were to suddenly be doubled. Would the instantaneous acceleration be given by.
2f(t)-b.(W)*-C (w)x 9
x=
a(w)
in general, no Or suppose the amplitude of the oscillation to be suddenly increased. Would the out of phase component of f(t), inediately after
the change, beequai to b*? Again, in general, no Thus, at best, b(w)
must be considered as a sort of "apparent" damping coefficient, a(w) is
an "apparent1' apparent mass, and the physical significance of both is obscure. When the oscillation consists Of several coupled modes, the
so-called coupling coefficients are equally confused and confusing.
If one restricts hImself to a phenomenologiCal investigation of how
a given ship behaves in a given wave system, these difficulties do not concern us. We simply measure responses to known waves. Most of the work
overtlie past decade has been of this nature, and much of it has been
exceilent However, sooner or later, we are required to consider not
"what,t'
but "why," and a mOre analytical technique is demanded. The phenomenological
study can tell us the effect of a change in ship loading on seakeeping
qualities only afterwe have measured it; there is no basis for quantitative
prediction given the results for one gyradius. And tie effect of achange in form i presented as an isolated result, unrelated and unrelatable to
the geometric parameters involved. We are driven to the use of the ôdél discussed
above,itl an attempt to clarify the relation of cause and effect. However; such a poor mirror of reality is of little value, and in fact can do mitch
hàt.
I am not the firstto raise this issue. . The difficulties are well
known, and a number of writers have discussed them. in particular, Tick has vigorouslY argued against our usual practice,
and has. proposed a model
which is very close to the one which will be exhibited here. His case is based solely upon the general characteristics of linear systems, while we Shall take advantage of the principles of hydrodynamiS to tie the model
to the phenomena.. More recently, Davis4 has proposed a rational approach particularly
since it was the spectral theory of statistics which at first gave weight
to the investigation of responses to periodic waves.
Briefly, the specific objectives of this paper are:
Th exhibit a model. which permits the representation of the
response of a ship (in ix degrees of freedom) to an
arbitrary forcing function. (with excitation in all six modes). The model. will not involve frequency dependent.parameters.
to separate the various factors governing the response into clearly identifible units, the effect of each to be separately
determinable. Thus the effect of gyradius will be separable
from added mass. The added mass will be related only to
inertial forces and moments. The nature of the damping force
will be exhibited. The effect of coupling will be d'rivable, and the effect of "tuning" upon coupling will e determinable.
In this paper we shall not consider the complementary prollem. of the relation of the exciting .force to the incident wave sstem. This probem
is equally basic, and when ithas been adequetely treated, we will begin to have a satisfactory framework for the interpretation of our empirical
THE IMPULSE RESPONSE FUNCTION
The basic tool whicI. will be used in this study is an elemcrtary one.,
widely used in other fields, and well known to all engineers; the imuIse
response function. It is difficult to understand jts neglect in our field.
Perhaps as Tick suggests, it is because waves look sinusoidal.
For any stable linear system, if R(t), the response to a unit impulse, is known, then the response of the system to an arbitrary otce
= (t-'r)f(T)dT
or
x(t) = R(1)f(t-T)dT
'Jo .
The only asSimptiofl required (aside from convergence) is linearity. El th
-present context this is, of course, a very strong assumption, and the purtests
will argue that it implies a thin ship or the equivalent? However all
experimental data indicate that the assumption is a good working approximation
for small to moderate oscillations of real ships fos We shall hyothesize
that.the assumption holds absolutely. V
Let Ci = 6) be displ4cements in the sx modes of response:
x1 = surge (positive forward)
= sway (positive to port) V
x = heave (positive upward)
x4 roll (positive, deck tO starboard) = pitch (positive, bow downward)
x6 = yaw (positive, bow to port)
Let R..(t) be the response in mode j to a uni-t impulse a
Note that R..() does not necessarily equal zero, though
13
which is not unstable, it will ordinarily be finite. In
f(t) is c [I]
[
L t t 0 in mode i.. in a dampe4 system: modes without aThus, the matrix.
IR .(t)I
completely characterises the response of the ship..1
4-?__
-to an arbitrary excitation.
,ILI
Ai-e
Before we go on, let us onsider the relation Of these functions to
the usual coefficients. First consider the case where the modes are
uncoupled. Let
'f(t) = F cos(Wt + e) (3]
where s' a phase angle whose value will be assigned later.
R(T) cos[(t-T))4T
=
F[cos(wt4)
cos uyrdr-i- sin (wt)
$
R sin WTdT]= [jc.
w) cos wt.) +
(w) sin (wt+cj)] (41where
(w) =
S0
COS'WT4T [..5a]and
(u)
= Rk. (i)
sin w'rdi
restoring force (sway, surge, and yaw) the impulse response will
asymptot-ically approach some va]ue. For other modes, Ri() = 0'.
If the [f(t)}are a set of arbitrary forcing functions, the corresponding
responses are
are the Fourier cosine and sine transforms of R.11(t). We shall call
these transforms the frequency response functions., We make the further reduct ion x(t) = F. [(RC. + R. sin .). cos wt
I
11 1 Taking = tan 1] ii Alsox(t) =
F.[(RY.)2
+R)2P
cos wt ) 11 f (t) cos wt - sin wt)F.
ccc,i
[8] i -[(R)2 +
(R)2]2
Now consider the usual representation
a.. + b.. + c.x. =
f(t)
1].
11
11
1[9]
Using the x and f. from [7] and [8], it is easily seen that
+ (Re. .
RC
in c.)
sin wt] 11l
1] 1 a.= 7 L 1
(RC ) + (RS ) ii ii R 11 R 11 =w[(R)
+ [lob]Jos2
£2?YZ_
we havetLt
-"A
J.j
[10 a]2
iT
Amore
useful relationship is obtained by setting = 0 in (4]:x(t)
....= (w). cos wt + Ri (w sin
i
Thus
Ri
and Ri ae the amplitudes of the in-phaseand
out-of-phase
components of the response to a unit amplitude
forcing
function of frequencyw. The
impulse
response fUnction s related to these functions byR(T)
$.Rj
(w) COswT:d1(j
(w)sin ur.
'dw 112]C S
using. the Fourier inversion formulas' Note that and R
are uniquely
related.. If one is
ktown,
then by [12]and
[5] the other is determined.Equation [11] can also be written
x (t) Where
R(w)
tan=
cR(w)
ThuS, the response fdl'lows the excitation
by the phase tan(R /RC) and
has thi amplitude [ (Ri)2
+ '(R) 1. .,
The response for a given frequency, as determined by the pair of
functions R
R Cor alternatively, the pair [(R8 )2 +
tan -
(R/R), is a
mapping in the frequency domain
of the unit responsefunction, which is defined in the time domain,. As equations' [4]
and [11].
permit us to pass from either domain to the other, the two representation,
are completely.'equivalent'. Viewed in' this way, the frequency response funct:ion is a meaningful, useful,' cOncept. It
is only when
we try to atTibute 'a deeper meaning to it, by imbedding it'in a false time..,domainmodel,.
that We create confusion.+
(R)2]
cos [wt - e(w][13]
23
Now consider the more genera1,cOuPlLesystem, with excitations in a
singlemode of the same form as given in Equation [3].
Then
x(t)
=F[RJ
cos (wt + sin (wt +[15]
If we consider the usual repreSeTttati0n
+ bjkij +
Cjkxj)
=where fk(t) = 0 for k i, we can develop a system
of equations in the
unknownS, aikl bik. (The cik are assumed
known from static measurements.)
All 72 of these
unknowns
are present, in principlesexcept where modes are
uncoupled. To determine them, it is necessarY
to consider the responses to excitations in each of the modes, separatelY.
We then have enough equations,
if we separate the in-phase and
out_of-phase components, to determine the coefficients. We have no need for
them
here, so we defer further discussion
until we
have
an analogous problem.It is only significant to note that they can, in principle be determined
from the set of impulse response functionS, and therefore they contain no
information which is not derivable
from these functions.
Setting = 0 in [15), we
have
the system= cos wt + siit wt
Thus, R and R5 are the amplitudes of the
in-phase
and
out-of-phase3responses in
the j mode to unit amplitude excitation
in
the i
mode.
As before, and 11..z\
RC coswtdw
a1(t)
ii
\
Rsinwtdw
=1.rlo.1.i
= +(R)21
cos (wt-e3) [16) 6 [18)Where
tan
RSJ/RJ
[19]
We
have
passed Over the question of convergence of the integrals inEquations [2] and [5]. Consistent with our hypothesis of linearity, we
shall asste
Jf(t)J
is bounded. There will then be no difficulty unless JJ'R1(T)d'I does not exist. Unfortunately, in three modes there are norestoring forces (or.else they are negative), and evidently some care is
needed in treating these cases.,, A negative restoring force implies an
unstable
system,
whichwould be beyond the scope of this analysis. However, the case inwhich
ápproaéhes some non-zerO but finite limit can betreated. The divergence of the integrals can be overcome, if we arbitrarily
assign a value to x(0) We can formally write
x(t)
=R(t-T)f(T)
dT )
R(-T)f(T) dr+
or
(t)
(T)fi(tT)
d+
£ [a(tT)
-
R.(T)]f(-T)
dT+ Xj(0)
[20]The second integral converges, .so this expression provides a usable
definition of x(t). Now let f(t) = cos wt. After an integration by
parts, We
have
x(t)
Sin w(-T)
dT£
[R(tT)
- R1 (fl cos u
dT +
(q)
Our only concern is with the oscillatory compànents of Xj. These are
easily determined by considering the asymptotic
form
of the above expressionas t becomes.large.
R(tT)
. and the second integral becomesthen
liin
5
x(0)
=
oLR(tT)
ax(t)
=;
-
Ri(T)J COSwtd
()
=j 'J
cos wt sin wt)where Ri and are the sine and cosine transforms of R.(t). We know
that x (t) is sinusoidal, with frequency w. Therefore, this expression
holds, not only, for large t, but for all t.
If we define
R1 =_R/w
[22a]R = [22b1
then [16] still holds. Note however, that and Ri are no longer trans-forms of Rij because these do not exist. Nevertheless, an inversion is
sill possible. Consider I
[R ('r)
- R()] cos
w1 dT1#0 ii
= - [Ri(T)_Rjj(a)I sin (UT -
!
R (T)sin wr dT
0
W'O
jj= -
=That is,
Ri is the cosine transform of [Ri(t) Rij()] andRij(t) =
R1() +
SO R cos wt dwLetting
t equal zero,RC dw
R(a)
=rr'o
ij
[21]
so
Ru (t) =
5
R (cos wt - 1) dwWhen R(o)
= 0, this reduces to [17a].Similarly,
[a(T)
R(cD)] sin wi- diI = -R (CD) /w + 30 Ru COS WT dr
= [Rc
- R (CD)]/w ii ii Ruj(t) = R(CD) [R18 ] sin WT dw CD R(CD) [1 2S
sin wi dw 1cR
sin wi dw = 0 W J W'10 ij CDsinwtdw
1T 0 ij since LCD sin wt w dw = IT 2Therefore, E17b] holds even when 0.
If R1 and Ri are
known,
it is not difficult to determine whether ornot R(CD) = 0. Equation [23] gives R&) in terms of R
[24]
CD
0
Also Tb Urn
2j
R5 sinwtdw
R(cx') 11 0jj
c1-urn al
asSflUtdw
t-'°rr'o ij w 1 im = = (O- 0 wRusing a well known theorem in Fourier transform theory (Reference 5, page 12).
2.5
When the matrix of impulse response functions is known, our firstobjective of finding a representation of the ship response which is free
of frequency dependence is achieved. These functions, which we shall
collectively call the impulse response matrix, can in principle be determined experimentally.. We shall discuss such an experiment, but we defer these
remarks until we have made some progress toward our second objective.
EQUATIONS OF MOTION
/ The transient response of a ship has been considered by Haskind,b who attempted an explicit solution of the. boundary value problem. This ,' we
shall not try, as we are concerned only with finding an appropriate form for the equations of motion to use as 'a basis for the interpretation of
experimental results. We do not agree with certain of Haskind's hypotheses,
and. our resulting equations differ from, his in several important respects.
Our approach is quite different, depending much more upon physical argument
than upon mathematical. analysis.. '' '. . . ...
Golovato7 carried out an experimental investigation; of the declining oscillation, motion in pitch.. We shall see later that this is equivalent to measuring the response. ,to an impulse However, Golôvato was not aware
of the equivalence between the transtent and Steady state responses which
we have. just. discussed,. so he attempted only to match. the coefficients derived from the transient expe;.iment at the frequency of the declining oscillation, with those front a forced oscillation experiment at this frequency. He was handicapped because of the anomalous behavior of the curve of declining amplitudes. For 'a simple harmonic' oscillator, this
curve isa straight line when plotted on semi-logarithmic paper. His
curves departed. radically from Such a pattern. Re recognized that this implied, that the mathematical model was faulty,. and attempted.; with some success, to fit his results. with forms based on Haskind's study.
More recently, Tasai8 has performed declining oscillation' experiments in heave., using two dimensional fOrms. His results are not significantly different from those of Gblovato. He matched his results at the 'measured frequency with. Ursell's theoretical results for forced oscillation. The agreement is quite good..
Just as the response of' a stable, linear, dynamic system to an arbitrary force' can be. given in terms.of the dynamic response to an impulsive force, the response. of our' hydrodyiiamic system can be stated in terms of the hydro-dynamic response to an' impulsive displacement. 3efore we demonstrate this
.1 principle in the most general case, let.us.verifyit in the simplest case
j, 2CA$E I NO FORWARD SPEED
Let-the ship be. fipating at rest in st.l water. We use a. system of
coordinates .(, ., ),fixed in space, with origin i the free surface
above the center..of graytty of.the .. ,.
At time t = 0, we suppose the ship to be given an impulsive displace-ment in the 1th mode, throug the displacement c1. Th time history .of
this impulse is not significant, but. for purposes of visualzatiofl, it may
be considered tconsiSt of a movement at a large, uniform, velocity V1 for a small timet, ith the motion terminated abruptly at the end of this
time interval. Then. ,.
AX1
v1At.
During the impulse, the flow will have a velocity potential which is proportional 'to the instataneOUs impulsive
therefore, be written where is a
flow..
will
satisfy the .condtions t#Pd& /(4_..4_ cs---'.
on Ca'i
.
4tOw
4i
/n
= 5j
onvelocity of the ship. It may,
normalized potential for impulsive
b1
jd3 )
= 0 [.26] whereSi =
' T i..= 1, 2,.3=r
a1j3
[.281 S [27]S = surface of the ship
= outwardly directed unit normal
-. th
i = unit vector in j direction
= position vector with respect to c.g. of ship
It is well known8 that the above problem is equivalent to that obtained by reflecting S in = 0, and taking the surface condition over the reflection
to be the negative of that over S. The solution to the Neuman problem for the flow outside this ste surface is also the solution to the given
problem in the lower half-space. For non-pathological Surfaces, the
solution exists, and in fact can be computed by means of modern, high-speed
equipment.1°
During the impulse, the free surface will be elevated by an amount
= - v t = - [29]
t.f
After the impulse, this elevation will'dissipate in a radiating disturbance of the free surface, until ultimately the fluid is again at rest in the
neighborhood of the ship. Let the velocity potential of this decaying wave motion. be qj(t)xj. It must satisfy the initial conditions
[30] and Xj = = -g
& on
= 0 orq1(Ci,,O,to)
= -g C3 [311Afterward, it satisfies the usual free surface condition,
2
+g
=0
and the boundary condition on S,
an
We may take this to hold on the original position of S, introducing errors
of higher order in
only.
This is a classical problem of the
Cauchy-Poisson type, and there exists an extensive literature on the subject.
With
condition
[331,
it is more difficult, by an order of magnitude, than the
Neumann problem.
Nevertheless, it has a well defined solution.
th
- Now
let the ship undergo an arbitrary small motion in the j
mode,
x(t).
To the first order, the velocity potential of the
resulting flow
will be simply
t
\
=
[34]
It is evident that the boundary condition on S
is satisfied on the equilibrium
position of S, as the first term provides the proper
normal velocity and
= 0 on this boundary.
But also, the value of
t3/an on the actual
posLtion of S will only differ from its value on S by terms
of second and
higher order in Xj and its derivatives, so we may consider that
[34] holds
on the actual position of the hull.
To verify that the free surface condition is satisfied,
first note
that
a2®d*
d*
acp1(0) = 4r +c(0)
+at
Xjçt.
a2cp1(t-T)
+ dTBy [261 and [30], on
= 0 this reduces to
[321
Also, or cp(0)
xi
+J
J at-!_tSt
cp1(t-T) i.(r) dT-
3Substituting these in the free surface condition
(
St
(
a21
) *.()
dT. = 0[35]
+
+g
by [311 and [32]. Thus, this condition is also satisfied, and 0 is the required potential.
The formula [34] is a hydrodynamic analog of [1]. It is quite general, and can for instance, be used to find the velocity potential due to a
sinusoidal oscillation with arbitrary frequency. It is, of course, necessary
to know the function
cp(t)1
and this presents unpleasant difficulties.In
this study we are content that cp(t) exists, and these difficulties do not
concern us.
Of more importance than the velocity potential is the force acting on
the body. The dynamic pressure, in our linearized -model,
is simply p = p t p
j j
Pj(0)Ij +S-ç1(t-T)
dTacp.(t-T)
=j$j +
(T) dT
[36] tThe net hydrodynamic force (or moment)acting on the hull in the k mode 2ç
Ct-i)
is then given by where I-,
ft
=£
d + p Sd
\
ç(tT)
sk
"-cp1(tT)£00
*(T)
d1£
Skd
d*j(r) dT
-;rHrjN'
= inertia of the ship in the th mode
th
= hydrostatic force in the k mode, due to
displace--' -' th
ment Xj in the .1 mode.
6jk = Kroneker delta 8jk =
1 if i
=k,
=0 if j
k)t
= m1 + (t-'r) (T) dT [37] wheremjk=PS*i
skda
[38]
(-z)
= do[39]
We can now write the equations of motion of the ship which is subjected to an arbitrary set of exciting forces, [fk(t)3. These will be
6
[(mjôjkmjkj
+ Cj lcXJ +
Kjk(tTxj(
dT]
= fk(t) [40] = Xj mjkCASE II - SHIP UNDERWAY
The case of the ship experiencing small oscillations about a reference position of mean uniform velocity is much more complex. A pair of functions,
arid p, no longer suffices, although the pattern of ouranalysis will be
similar to that followed in Case I.
We use a fixed reference system, with = 0 on the free surface and with the c.g. Of the Ship at = 0 at time t 0. We suppose the ship to
be
moving
with a uniformvelocityV in the direction.Consider the Cauchy -. Poisson problem defined by [30], [31], [32], and [331, except that now [33] is to hold on the moving surface S. This problem has a solution cp1(C1,C2,,t,v) which is, of course, 1denical withthe Ccj
of Case I when V = 0. USing this
q'j and the obtained in Case I, we may
write the velocity potential for steady motion,
where
V1(C1vt,C2,C3) +
$T,t-T
dT] [41]= c
(Ci-VT,C2 ,C3 ,t-T)That this satisfies the boundary condition on S is evident, as Vi1 provides the necessary instan;aneous normal velocities, and j/n =0 on S for' all.
. The free surface condition :s also satisfied.,'as
may be verified by
direct evaluation,, as in Case I..
The velocity potential for tbe flow generated by the. ship moving with
constant velocity, after an impulsive start at time zero, is
v [ui (Cj-Vt,C
S0
cp ('r,t-T) dl] [421The free surface and ship surface conditions are satisfied as before. The surface elevation at t = 0 is '
or
*fj
j n =
i1 -
on S (displaced)In three cases, solutions are immediately available. If j = 1
4. (C.-txi ,12 ,iIia) = 4r-xi
+ O(x1)
is a solution of [431 and [441, since in this case we have simple translation. Therefore [44] [44a] [45a] [
=![..V.+,(O,O)]=O
as required, and the initial conditions are met. Therefore, this must be
the stated potential.
We shall need the steady motion velocity potential for the case in which the ship is displaced by Xj from its reference position. We could,
of course, consider the displaàed ship as a completely new hull, and write down a potential similar to [41], with new functions $ and q.. Instead, we determine the corrections to the i and çj discussed above which ae necessary to satisfy the new boundary conditions. We wish a such that
*1 +
Xj = 0 on Ca = 0 [43]which implies that
lj = 0 [43a]
Also
-
Similarly
or
so
II,
y12-
-For j = 3, there is no such simple solution. Noting that the right side of [44a) is zero on S(original), it is only necessary to find its change
when S is displaced. Then
*i3
[45b]
*i3
-
nC3
[46]If j = 6, the displacement is simply a rotation in yaw. The
trans-lation of a yawed body is equivalent to simultaneous transtrans-lations parallel
and perpendicular to the body axis. .Theref ore, the solution to [43] and [44] is
-(
4'=
*1 +
-
Ci+ *2)
sie = C
-
Ci + 1P [471If .j = 4, or 5, the first
term
of [44a] becomes(T3x)]
iiThe second term is
which may be written, using values of and 7* evaluated on S (original),
-
t74ri+x1(i_3Xr.7)74?i)
If we drop terms of higher order in Lx
and
use 1:27], condition [44a] reduces to .4-.
-4 -4-
ij_3 Xfl
i][i3Xn.7*i+n(i_3X7)71Pil
or andr ,'
si
2*i
2'Pi 1
=-
[n.i
- -T
- Ci [48b]Conditions [43a) and
[44a]
are sufficient to determine Strictly,[a1
holds on S (displaced), but we only introduce errors of order (ax)2 if we take the ship surface condition to hold on the original S. Similarly, [46] and [48) can be applied on the reference position of S.To corresponds a with
=g
forCs=0,t=0
[49]and with conditions corresponding to
[301, [32),
and [33] holding.Again
we take the ship surface condition to hold on the reference position of S.We need yet one more pair of functions. The
normal
derivative /rzwill
differ from zero on S (displaced), to the first order in AX To correct it, we define a function which satisfies the conditionsol
= _._j.
rt
cp(T,t-T)
dT on S(displaced) [50]4
r
(
1I1i a*1=
-
I
and
4t 0 on
C3=0
oj
As we do not intend to exhibit solutions for
P,
we shall not reducethe right side. We also need a
oj' with
Poj
=
and the other appropriate conditions also holding.
We now have all the pieces needed to write the velocity potential for the flow about the ship when displaced by from its reference position.
It will be 1(T,t-T) dT] 8 = (T,tT) dT] + Ax t + 'j
{oj
q,0(T,tT) dT]}
The 'ms
v($j +AX1)
provide the necessary normaj. velocity in the displaced position. The normal
velocities due to
rt
v tx
oJ and V)_a, (T,t-T) dT
cancel, and none of the other terms contribute normal velocities of first order in &. Therefore, the. ship surface condition is satisfied in the
displaced position. Further, each pair of terms in brackets satisfies the
free surface condition, as may be verified by direct evaluation.
We also have all the pieces needed to assemble the potential for the
flow generated by a ship experiencing small oscillatiorm[xj(t)3. It will be
=
{ [+
(,t-)
dT]
[Xjlj+$
dT]
t
dT]}
L
j
cp(Tt-T)*(T)
[54]The ship surface condition is satisfied as before, except now the term
provides the additional, components required for the oscillatory velocities. And again, the bracketed pairs of terms satisfy the free
surface condition.
The dynamic pres.sure at any point in the fluid is given by
6 ol
j
CiCi
I]
-
_Y
-
x
___
p -
t
t
Pcp1(TtT)
(rtT)
]
dT(It
cp1(T,tT)
}
+ dT 1 1cpx(T,tr)
V2 i_co tdT)
[55]There are two convolution integrals in [55], one involving the
oscillatory displacement and one involving the oscillatory velocity. These
may be reduced to one by means of an integration by parts. We can go
so that
CT01cpj1
cp41 L +-r11
cp01 1(0)
1
L+
Jx(i-)
dT =x(t)
J_ci, O)_$
(t_T)
='?
Ci-Equation [55] now reduces to
+
_.t"
+(
j
oJi)
x
V2j
oci
+S:(cP(Tt_T)
- V
(t-'r)
)
j
(.)
dT}
C
1çt
vpj(T,t-T)
dT)
are concerned with the oscillatory value of the hydrodynamic
not steady components.
The last term [59] does not involve the
However, when we integrate the pressure over S, the fact that
Sp
(i) dT
[60]
force, but
[XJ).
is changing
T[cp(Tt-T) + cp(Tt-T)] dT
= cZ(tT0)
[56]
The significance of this function cD
can be seen by rewriting the potential
f or the uniform flow with the body deflected (Equation [53)).
It becomes
{
+(,t-) dT+&[
[59)
and
t
[57]
dT[58j
its position in a steady flow field implies that even thi1s. term contributes to the oscillatory pressure. These pressures will be functions of the displacement, only.
Integrating the pressure over the surface of the ship., we can write the equation of motion:
th
c1x4 = Total hydrodynamic and hydrostatic force in the k mode,
JL%J th
due to displacement in the .J mode.
Kjk(t_T) =
C(
acp.(r,tr) at
There are symmetries which reduce the number of coefficients. For instance
m.k = P .Sk d
C
= -p
Itj-
doIf we consider the space enclosed by S, the free surface, and an infinite
hemisphere, we can apply Greent s theorem, and we find
mJk=-PS
an!i. do =Further, if we consider the transverse symmetry of the ship, the matrix a.(t-T) \\
-v
[63] [.64]il
[(mJoJk+mjkj
+ bjk*j + CjkXj +K(t-T)
*j(T)
dT] = fk(t) [61] where m. and mk are as defined in [40] and [381, and
.3 j
S5
($ljoj
)
5k d [621Evidently, the matrix [bik) is of the same form, except that in general
bik # bkj. The matrix cik is even simpler, as surge and sway displacements provide no restoring forces, hydrostatic or hydrodynamic.
Therefore
The nlatrix[kJk(t)) is of the same form as [bik)
Equations [61], though similar in form to these developed by Haskind, aaskind found no hydrodynamic force proportional to the displacement, nor did he find the components of b due to and He also found that b33 = b = 0, and b = - b. The presence of in the definition of
bik makes it unlikely that such relations hold here. Further, his kernal
in the convolution integral must differ from that found here. The reason for these differences is that Haskind neglected terms in satisfying the
oundary condition on the displaced S which are of 4rst order in
Xj.
With equation [61], we have advanced a long way toward the second
objective of this paper. The dynamics of the body have been separated from
the dynamics of the fluid. Further, the hydrodynamic effects have been
[Cik) = 0
0
C31 0 c51 0 0 0 0 C43 o C820
0 C33 0 c53 0 0 0 0 C44 0 C64 00
C350
c55 0 0 0 0 C46 0 C68 [66] jk3 reduces to [miki = m11 o i o '115i O 0 0 1fl42 0 11152 11113 0 fl15 0 11153 0 0fl4
0
11144 0 11154 11115 0 D15 0 fl15 0 0 0 111 0 [651separated into separate, well defined, components, each of which can be found (in principle) from the solution of a Neumann problem or a Cauchy
-Poisson problem. Specifically, we draw the conclusions:
The equations of motion are universally valid, within the
range of validity of our assumption of linearity. That is
any excitation, periodic or non-periodic, continuous or discontinuous, is permissible, just so it results in small displacements from a condition of uniform forward velocity. The case of motion with a negative restoring force, or at
least the early history of such motion, is not excluded.
The inertial properties of the fluid are reflected in the
products m.kx.. The coefficients are independent of frequency
and of the past history of the motion, so they are legitimate
added masses. Further, they are independent of forward velocity.
There is an effect proportional to which accounts for some
of the damping. This effect 'anishes when the mean forward
speed is zero.
)
There is a hydrodynamic "restoring" force (it may be negative). It is equal to the difference between the hydrodynamic forces acting on the ship due to the steady flow in the equilibruim position and the deflected position
The effect of past history is embedded in a convolution integral over *(t). For sinusoidal motions, this integral will ordinarily have components both in phase with the motion
and 900 out of phase. The latter component contributes to the damping.
L71. HYDRODYNAMICS OF THE IMPULSE RESPONSE FUNCTION
We now have two systems of relations between the excitationand the response of the ship, the impulse response relations, [2], and the
equa-tions of motion, [61]. The former are of greater value in describing the response to a given excitation, while the latter are useful in analysing
the nature of the response. Both systems hold for small oscillatory
mo-tions, so there are relations between them. We shall examine these.
First, let us start with the equations of motion, and derive the
functions fR (t)i. Suppose a ship, moving at constant forward velocity, to be subjected to a unit impulse in the I mode at time t = 0. During
the impulse, the equations of motion reduce to
k = i 6ii
where 8k is the Kroneker delta. Suppose the impulse acts during time
tt. Then, since c.
t=b*.=R (+0)
3 3 y we have 6 "k Rik(+ 0) +m R
jk ij (+ 0) 8ikAs i and j range independently from 1 to 6, we have 36 equations relating
the two sets,
fmI
and [R..(0)1. If the equations of motion are known,equations [671 fix the initial conditions from which the impulse response
functions can be determined. Conversely, if the impulse response func-tions are known, these equafunc-tions yield the apparent masses.
Immediately after the impulse, we have
and = 0(t) = x (0) + 0(t) 3 j x. = (0) + 0(t) 3 j K(T)*(t_T) dT = 0(t) 0
Therefore, considering only zero order terms in t, the equations of motion
yield:
R() +1[mik.cJ() + bik
R1(+0)] = 0 [68]which relates the coefficients [b.k} to the accelerations [R(+0).
Now suppose the. ship to acted upon by a constant unit force in the jth
mode (we assume a positive restoring force to exist in this mode).
Then, after equilibruim is reached,
6
Y
= 6ikjl
=SORij
(r)dT or 6R(T)
dT = 6ik [69] j=lIn modes without a positive restoring force there is difficulty, as there
is no guarantee that all of the coupling coefficients are necessarily zero.
Thus, c x
,
the sway force due to a yaw angle x6 will not ordinarily be
If we rewrite [61] in the form
6t
$iccr) R1. (t-T)dT = [70] j=l 6-
[mj
6jk + mjk) i(t) + bjkRjj(t) + Cjk1jj(t)] j=lwe have a Bet of 36 equations which can either be regarded as a set of
simultaneous integral equations for the kernals fICjk(T)} or a set of simultaneous integro-differential equations for the impulse response
functions, fR(t)3.
We have already seen (equation
[16])
that iff1(t) = COB (Vt
then
xj(t) = R COB Wt + R sin (Vt
Substituting these values in the equations of motion, we get
+ mjk)UPRIJ - bik
WRjJ -
cjRjj
cu(R K + R
K)]
(Vt+
[mj
8jk + mJk)uPRi - bik T.-
Cik R- w (R - R1
K)]
sinPor any given frequency, this is an identity, so the net
coefficients of
cos wt and sin wt must be zero.
This gives us 72 equations relating the transforms IRi.,
R1.3
with the transformsKjk}. We have, or, equivalently, 6
{[c
6Jk+mjk)W_cjk_wKfk]R
C S-
(bik + Kjk)w Rij} = 6ik S C C Bcn)(R Kik +
Kik) 6
C = 6ik [(mj
8Jk + mk)aR
- bJkw R - CikRj]
j=l 6 r-i C C S S
-
w' (R Kik - R1 Kik) /- ij j=l 6= w
jk + mjk)ciP R + b U) R C -cjkRij jk ij j=l Y{bJk+ K) U) Rjj+
[mj
6jk + mjk)U? - Cjk_ K;klRij} = 0 [72bJThus, instead of the integral and integro-differential equations relating [R) with [KjkI Equation [70],we have systems of linear
equations relating their transforms.
The transforms of [RJ also yield useful variants of the relations
already given. For instance, if we let U) = 0, we have
[71a1
[71b]
and we have =
4Cw
R1 dwi(0) =-
_LJ
°?
RC dw TI o and 6Cj R(0)
= 6ik .1=1a more general form of 169].
Also, noting that
* (t) =
__.j'
W R (w) cos wt dw .1ir0
ii(t) =-
uP iT JoR(W)
COB wt dwTherefore, [67] and (68] may be written
6
8jk +
mjk)5
w R dw][m
8jk +
j=1 rr- T °ik
[73] (74a] [74b] [75] [76]0
Rj dw - bjk$w R
dW] = 0'5
CONCLUSIONIn the foregoing, we have presented two mathematical models for rep-resenting the response characteristics of ship. The equations of motion
are more general, as they apply to the initial stages of an unstable motion. Where the two systems are equally valid, we have relations which permit us
to pass (at least in principle) from either system to the other.
The impulse response function is certainly the better representation
for computing responses. It integrates all factors, mechanical,
hydro-static, and hydrodynamic, in the most efficient manner possible for corn-utation. However, for this very reason, it is a poor analytical tool for explaining why the ship responds the way it does, or how the response will be affected if any change in conditions occurs. For instance, models
are ordinarily tested with restraints in certain modes. A restraint in any mode will affect the impulse response function in ay coupled mode. Since the ship Is free in all modes, itis evidently imprto use these response functiotto predict full scale behavior, unless they are corrected
for the effect of such restraints.
The hydrodynamic equations do not suffer from this disadvantage. Known restraints are readily includable, and their effects determinable. Or a change in mass distribution can be treated independently of the
hydrodynamics.
It is not uncommon in model testing to
have "incompatiblet' parasitic inertias in the different modes. Thus, the towing gear may con-tribute a different mass in surge from that in heave. By means of the equations of motion, the effect of these inertias upon the motions can beanalyzed. Thus, the equations of motion provide a more powerfulanalytical tool for studying the relationship of the response to the parameters
governing that response.
We can conclude, then, that these two representations complement each other; the one for response calculation, the other for response analysis. In fact, if it is truly practicable to pass from one representation to the
a) Model experiments may be designed to obtain maximum accuracy,
rather than max1miim realism. Hydrodynamic effects should be emphasized in the design, since other effects are separately
determinable. Thus, one.should test at small gyradius, in order that the effect of the inertial properties of the body
itself will be minimized.
Restraints are permissible, if their character is fully known. Thus, rather than directly find the impulse response matrix, in its complete generality, more elementary experiments may be conducted to determine specific terms in the equations of
motion. One may restrict himself to one, two, or three degrees of freedom, and obtain results which are completely valid, when
interpreted by means of the equations of motion.
The recurring difficulty of handling modes in which the restoring force is zero or negative can be easily overcome. It is clear that an accurate experimental investigation of these modes would uncover practical difficulties analogous
to the theoretical ones we have discussed. However, the problem can easily be solved by imposing known restraints
(i.e. springs) which will restore positive stability. The effect of these restraints is readily includable in the equations of motion, it can be removed by calculation, and
the correct impulse response, free of restraint, can be determined.
REFERENCES
Weinbium, G., and St.Denis, Manley, "On the Motions of Ships at
Sea," Transactions, The Society of Naval Architects and Marine Engineers, Vol 58, 1950.
St. Denis, Manley, and Pierson, W. J., Jr., "On the Motions of Ships in Confused Seas," Trasactions, The Society of Naval Architects and Marine Engineers, Vol. 61, 1953.
TIck, Leo J., "Differential Equations with Frequency-Dependent
Coefficients," Journal of
Ship
Research, Vol. 3, No. 2, October 1959.Davis, Michael C., "Analysis and Control of
Ship
Motion in a Random Seaway," M.S. Thesis, Massachusetts Institute of Technology, June 1961.Sneddon, Ian. N., "Fourier Transforms," McGraw-Hill Book Company, Inc., 1951.
Haskind, M. D., "Oscillation of a Ship on a Calm Sea," Bulletin
de l'Academie des Sciences de l'UR.SS, Classe des Sciences Techniques, 1946 no. 1, pp 23-34.
Gblovato, .P., "A Study of the Transient Pitching Oscillations of a Ship," Journal of Ship Research, Vol. 2, No. 4, March 1959.
L
Tasai, Fukuzo, "On the Free Heaving of a Cylinder Floating on the Surface of a Fluid," Reports of Research Institute for Applied Mechanics,
Vol. VIII, No. 32, 1960.
Weinblum, G. P., "On Hydrodynamic Masses," David Taylor Model
Basin Report 809, April 1952.
Smith, A. M. 0., and Pierce, Jesse, "Exact Solution of the Neumann Problem. Calculation of Non-Circulatory Plane and Axially Symmetric Flows
AEODY1AMI
.W.E. Cunirnins
IMPULSE RESPONSE FUNCTION AND SHI PMOTIONS.
S(.1PSTEKN4S( FOF$TTUTT
I(ORTFøR7"
DA1V MO11AThLISTEFØRT:/
HYDROMECHANICS LABORATORY RESEARCH AND DEVELOPMENT REPORT
THE IMPULSE RESPONSE FUNCTION AND
SHIP MOTIONS
by
W.E. Cummins
This paper was presented at the Symposium on
Ship Thecry at the Institut fiir Schiffbau der Universitt
Hamburg, 2 5-27 January 1962.
ABSTRACT
''
After a review of the deficiencies of the usual
equat.ions of motion for anoscillating ship, two new representations are given. One makes use of the impulse
response function and depends only upon the system being linear. The response
isgiven as a convolution integral over the past history of the exciting force with the
impulse response function appearing as the kernel. The second representation is
based upon a hydrodynarnic study, and new forms for the equations of motion are
exhibited. The equations resemble the usual equations, with the addition of
con-volution integrals over the past history of the velocity. However, the coefficients
in these new equations are independent of frequency, as are the kernel functions
in the convolution integrals. Both representations are quite general and apply to
transient motions as well as periodic. The relations between the two
representa-tions are given. The treatment considers six degrees of freedom, with linear
coupling between the various modes.
The Impulse Response Function and Ship Motions
W. E. Cuinmina
Introduction
Just over a decade ago, Weinbium and St. Denis) presented a comprehensive review of the -state of knowledge at the end of what we may call the "classical" period in reaeardt on
sea-keeping. Soon after, St. Denis and Pierson2) opened the
"modern" period (some would prefer to call it the "statistical" period). The studies of the former period were primarily con-cerned with sinusoidal responses to sinusoidal waves, but the
introduction of speceral tediniques opened the door for the
discussion of responses to random waves, both long and short
crested. The construction of the spectral theory on regular ways they as a foundation delighted us all, asit presented an apparent justification for the admittedly attificial studies of the "ilassical" period.
The activity during this last decade has been spectacular, with five major and many minor facilities for seakeeping re-seardi being opened. Hundreds of models have been tested,
many full scale trials have been run, and there has even been some real growth in our knowledge of the subject. In particu-lar, the spectral tool has been sharpened and tempered by the
empiricists, and the analysts have made important advances with the rather frightful boundary value problem. In fact, we
have aU been forging ahead so rapidly that we appear to have
forgotten that we are wearing a shoe whidi doesn't quite fit. The occasional pain from a misplaced toe is ignored in our
general enthusiasm for progress.
The "shoe" to whidi refer is our mathematical mode], the
forced representation of the ship response by a system of
second order differential equations. The shoe is squeesed on,
with no regard for the shape of the foot. The inadequacy of the shoe is evident in the distortions it must take if it is to be worn at all. I am referring, of course, to the frequency
de-pendent coefficients whith permit the mathematical model to
fit the physical model (if the excitation is purely sinusoidal,
that ía).
But what happens when we don't have a well defined
frequency? The mathematical model becomes almost meaning-less. True, a Fourier analysis of the exciting force (or
encoun-tered wave) permits the model to he retained, but physical reality is almost lost in the inflaLty of equations required to
represent the motion.
Let us consider this mathematical model briefly, and restrict ourselves to a single degree of freedom. To be completely fair, let us consider a pure, sinusoidal oscillation. The forcing
func-tion (if the systemic linear) will be sinusoidal, and can be
broken into two components, one in phase with the displace-ment and one 90° out of phase. We further divide the in-phase component into a restoring force, proportional to the
displace-ment, and a remainder. The latter we call the inertial force, and treat it as U it were proportional to the instantaneous
1) References are listed at tim end of the paper
acceleration. The out-of-phase component, whith provides all
the damping, we treat as if it were proportional to the
in-stantaneous velocity.
We can now write an equation, whidt has the appearance of a differential equation, relating these various quantities:
a(w) +b(w)± + c(to)x F08in(un+ a). But a differential equation is supposed to relate the instan-taneous values of the functions involved. If the periodic
motion continues, this condition is satisfied. Of course, it could just as well be satisfied by the equation
ha + (c - an2) x = f (t)
or more generally
(a±d)+b*+(c+dw2)x=i(t)
where d is arbitrary. These are all equally valid models. One of them isto be preferred only if it truly relates the displace-ment and its first and second derivatives to the excitation in
some more general way. But suppose f (t) were to be.auddenly doubled. Would the instantaneous acceleration be given by
2f(t)b()ic(w)x
a (to)
In general, no! Or suppose the amplitude of the oscillation to be suddenly increased. Would the out of phase component of
f(t), immediately after the diange, be equal to ha? Again, in genera], no. Thus, at best, b (to) must be considered as a sort of "apparent" damping coefficient, a (to) as an "apparent" apparent mass, and the physical significance of both is obscure. When the osi4lInHn consists of several coupled modes, the so-called coupling coefficients axe equally
con-fused and confusing.
If we restrict ourselves to a phenoinenological investigation of how a given ship behaves in a given wave system, these
dif-ficulties do not concern us. We simply measure responses to
known waves. Moat of the work over the past decade has been
of this nature, end much of it has been excellent. However, sooner or later, we are required to consider not "whet" but "why," and a more analytical technique is demanded. The phenomenological study can tell us the effect of a change in
ship loading on seakeeping qualities only after we have
mea-sured it; there is no basis for quantitative prediction given
th results for one gyradius. And the effect of a change in form
is presented as an isolated result, unrelated and unrelatable to the geometric parameters involved. We are driven to the use of the model discussed above in an attempt to clarify the relation of cause and effect. But audi a poor mirror of reality
is of little value, and in fact can do much harm.
I am not the first to raise this issue. The difficulties are well known and a number of writers have discussed them. In parti-cular, 'fleha) has vigorously argued against our usual practice and has proposed a model which is very close to the one which will be exhibited here. His case is based solely upon the
vantage of the principles of hydrodynamics to tic the model
to the phenomena. More recently, Davis) has proposed a
ratio-nal approach from the point of view of statistics. This is sug-gestive, particularly since it was the spectral theory of
stati-stics which first gave weight to the investigation of responses to periodic waves.
Briefly, the specific objectives of this paper are:
To exhibit a model which permits the representation of the response of a ship (in six degree of freedom) to an arbitrary forcing function (with excitation in all six
modes). The model will not involve frequency dependent parameters.
To separate the various factors governing the response into clearly identifiable units, the effect of eadi to be
separately determinable. Thus the effect of gyradius will
be separable from added mass. The added mass will be related only to inertial forces and moments. The nature of the damping force will be exhibited. The effect of coupling will be derivable and the effect of "tuning"
upon coupling will be determinable.
In this paper we shall not consider the complementary pro-blezn of the relation of the exciting force to the incident wave
system. This problem is equally basic, and when it has been
adequately treated, we will begin to have a satisfactory
frame-work for the interpretation of our empirical studies.
The Impulse Response Function
The basic tool which will be used in this study is an
elemen-tary one, widely used in other fields and well known to all engineers: the impulse response function. It is difficult to understand its neglect in our field. Perhaps as Tick suggests,
it is because waves look sinusoidal.
For any stable linear system, if R (t), the response to a unit
impulse, is known, then the response of the system to an arbi-trary force f (r) is t x (t) = j' R (t - v) I (c) dv -or (1) x (t) =fR (r) f (t- v) dv.
The only assumption required (aside from convergence) is linearity. In the present context this is, of course, aery strong assumption, and the purists will argue that it implies a thin
ship or the equivalent. However all experimental data indicate that the assumption is a good working approximation for small
to moderate oscillations of real ship forms. We shall
hypo-thesize that the assumption holds absolutely.
Let Xj, (I = 1,...., 6) be displacements in the aix modes
of response:
= surge (positive forward)
x2 = sway (positive to port) = heave (positive upward) x4 = roll (positive, deck to starboard) x5 = pitch (positive, bow downward)
Xe yaw (positive, bow to port)
Let Rij (t) be tho response In mode j to a unit impulse at t = (1
in mode L Note that R1 (00) does not necessarily equal zero,
though in a damped system which is not unstable, it will
ordi-narily be finite. In modes without a restoring force (sway, surge, and yaw), the impulse response will asymptotically approach some value. For other modes, R (00) = 0.
If the {fi (r)} are an arbitrary set for forcing functions, the
corresponding responses are
2
Before we go on, let us consider the relation of these
func-tions to the usual coefficients. First consider the case where
the modes arc uncoupled. Let
fi(t) = Ficos(wt + ci) (3]
where ci is a phase angle whose value will be assigned later.
xt(t) = Ft j'Rii(e) cos(w(tt) + e)d't
= F1 [cos (WI + Ct)SRn cos wvdc
0 + sin (en + ci) j'Rt sin covdv]
= Fi (R11c (Ii)) cos (wt + et) + Ru! (w) sin (on + es))
(4]
where
where
5 =
____
Rite (m) = $ R (v) cos ortdr (5a]
Ru5 ((0)= SR (e) sin wvdt [5b) are the Fourier cosine and sine transforms of (t). We shall call these transforms the frequency response functions. We make the further reduction
x(t) = F1[(R110cosE1 + R115sine1) coatst
+ (R115 cos c, - R11' cos Li) sin ofl]
Taking tan B1 = Rn5/Ri,C (6]
we have x1 (t) = F1 [(R)2 +R11c)!J'/5 cos cat. [7]
Also
f (t) F1 (R11t cos cut R116 sin cat) (8)
((R1j)2 + (R11C)n]'/S
Now consider the usual representation
a11 + b1; + c1; = f1(t) - [91
Using the x1 and f1 from (7] and [8], it.Is easily seen that
I R-. 1
a1 = i/wI c1 " I liOn)
[ (R)t + (R11)5j
IlOb] 0) [(R11c)! + (R.1')2)
A more useful relationship is obtained by setting e1 =0 in (4):
= R(o) cos cot + Rjj(w) sincot - (11]
Thus R,c and Ru5 are the amplitudes of the in-phase and out-of-phase components of the response to a unit amplitude
forc-ing function of frequency ca. The impulse response function
is related to these functions by
2('
R (v) - 1 R1(u) coscar de
0
__$Rii5(O3)5in(01CdO) [12]
using the Fourier inversion formulas. Not that R11 c and R116
are uniquely related. If one is known, then by [[2] and [51,
the other is determined.
Equation [11) can also be written
R112 + (Rj1)91' cos (cat - a1 (cc)] (13]
tan E = Re5 (w) [14)
R110(co)
xj (t) = $ R11 (r)fi (I - v) dc. [2) i1 0
Thus, the matrix {R (t)} completely characterizes the response Thus, the response follows the excitation by the phase
The response for a given frequency, as determined by the
pair of functions R.c, or alternatively, the pair [(R1)2 + (R)5J", tan4 (R,1/R1c), is a mapping in the frequency do-main of the unit response function, whidi is defined in the
time domain. As equations 141 and [11] permit us to pass from
either domain to the other, the two representations are
com-pletely equivalent. Viewed in this way, the frequency response
function is a meaningful, useful concept- It is only when we try to attribute a deeper meaning to it, by imbedding it in a
false time domain model., that we create confusion.
Now consider the more general, coupled system, with exci-tations in a single mode of the same form as given in equation
131. Then
xj (t) = F1 [R3 cos (cot + e1) + R,5 sin (cot + a1)). (15)
If we consider the usual representation
a
(a1 + bjkij+C3kXJ) =
and
I '
where k (t) 0 for k # i, we can develop a system of equa-tions in the unknowns, sJI, b - (The 0jk are assumed known
from static measurements.) All 72 of these unknowns are present, in principle, except where modes are uncoupled. To determine them, it is necessary to consider the responses to excitations in each of the modes separately. We then have
enough equations, if we separate the in.phase and out-of-phase
components, to determine the coefficients. We have no need for them here, so we defer further discussion until we face a closely related problem. It is only significant to note that they can, in principle, be determined from the set of impulse
response functions, and therefore they contain no information which is not derivable from these functions.
SettingE 0 in (15), we have the system
x1(t)
Rjj COBcot + R38 sin cot. (16)
Thus, R1 and R115 are the amplitudes of the in-phase and out.of-phase responses in the j mode to unit amplitude
ex-citation in the th mode. As before,
R11 (t) =
1..
R.11C cos Cot dts= R5sincotdw
= [(Rc)z + (R1B)9'hl cos (cot - a)
where tanej =
We have passed over the question of convergence of the inie-grals in equations [21 and [51- Consistent with our hypothesis of linearity, we shall assume f1 (t) us bounded. There will then be no difficulty unless flR13 ( dt does not existS Unfortunate-ly, in three modes there are no restoring forces (or else they are negative), and evidently some care is needed in treating these cases. A negative restoring force implies an unstable system. which would be beyond the scope of this analysis. However, the case in which R11 approaches some non-zero but finite limit
can be treated. The divergence of the integrals can be over-come if we arbitrarily assign a value to x, (0). We formally
write t 0 xj(t) = fR11 (t_t) f1(t) dtSRu (.c) f1 (-c) dz+xj (0) °'xj(t) = $R11 (-a) f1 (tt) d 3 + $[R1 (t + t) Ru (t)] f() dt + xj (0.) (20]
The second integral converges, so this expression provides a
usable definition of xj Ct). Now let f, (t) = cos cot. After an inte-gration 1,y parts, we have
; (t) = 1/co 5 1L (t) siflW(tT) dr
+ 5 [R1 (t + a) - Ru (a)) cot we dt + ; (0). Our only concern is with the oscillatory components of x1.
These are easily determined by considering the asymptotic form of the above expression as t becomes large. Ru (t+t)+Rjj(00), and the second integral becomes constant. if we set
x1 (0) =
-
jRjj (t + a) Ru (t)] cos cot dcthen
Xj(t) =
-.! (kf
cos cot + Rjcsinwt) (21]where ItJ and are the sine and cosine transforms of
u (t). We know that x (t) is sinusoidal, with frequency U).
Therefore, this expression holds not only for large t but for all t
If we define
Re = -
[22a)(22b1
then [16] still holds. Note however, that Rue and R,f are no
longer transforms of R1 because these do not exist. Neverthe-less, an inversion is still possible. Consider
j'[Rjj (*R(°°))cos wtda
=Rjj(a)_Ru(00)Isinwtl_--- kij(t)sinoYedt
0) C
0
Ru0.
That is, R110 is the cosine transform of (R1 (t) Ru (00)] and
R11(t} = R11(co) + - .$Rticcoswtdw
Letting equal zero,
-(23]
so
Ru(t) =
_L$Rif(coswt_1)dw
[24]When R,1 (00) = 0, this reduces to [17a). Similarly,
35 [Ru (-a) - Ru (00)] sin err d-a
= R11 (00)1w + $ ucos Wt dt
0)
0
= (1111C_..R,j(oO)]/w and
R(t) = Ru(00) +
_L$[Rs
Ruoo)1sin cot dco0)
0
[ha)
(17b1